Artificial intelligence based generation of synthetic user profiles

Abstract

An online system generates synthetic user profiles of users. The system receives a trajectory representing variation of relevance scores of epochs. The trajectory may be specified as one of: an image comprising a curve representing the trajectory, a set of tuples representing a graph of the trajectory, or a natural language description of the trajectory. The system generates a user profile comprising a sequence of epochs such that the relevance scores of the sequence of epochs varies over time according to the specified trajectory. The synthetic user profiles generated by the system may be used for training a machine learning based model or evaluating a system processing user profile information.

What is claimed is: 
     
         1 . A computer-implemented method comprising:
 receiving a request to generate a synthetic user profile comprising a sequence of epochs, each epoch associated with a score evaluating the epoch, wherein the score is determined by executing of a set of instructions that evaluate a set of events associated with a hypothetical user that occurred during a time period, the request specifying a trajectory representing variation of scores of epochs of the sequence of epochs over time;   generating the synthetic user profile, comprising:
 generating a prompt requesting a machine learning-based language model to generate the synthetic user profile, wherein the prompt specifies a set of parameters including a representation of the trajectory specified in the request, 
 sending the prompt to the machine learning-based language model; and 
 receiving a response generated by executing the machine learning-based language model, the response comprising a sequence of synthesized epochs, wherein each synthesized epoch has synthesized events such that the scores of the sequence of synthesized epochs varies over time according to the trajectory specified in the request; and 
   providing the synthetic user profile as input to a target system.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein the trajectory is specified as an image comprising a curve, the computer-implemented method further comprising:
 specifying the image representing the trajectory as part of the prompt generated for providing to the machine learning-based language model.   
     
     
         3 . The computer-implemented method of  claim 1 , wherein the trajectory is specified as a sequence of tuples, each tuple comprising an x-coordinate value and a y-coordinate value, the computer-implemented method further comprising:
 determining an expected score for each epoch of the sequence of epochs based on one or more of interpolation or extrapolation of values from the sequence of tuples; and   specifying the expected score for each epoch in the prompt generated for the machine learning-based language model.   
     
     
         4 . The computer-implemented method of  claim 1 , wherein the trajectory is specified using a natural language description of the variation of values of the scores over time, the computer-implemented method further comprising:
 including the natural language description of the variation of values of the scores corresponding to the trajectory in the prompt generated for the machine learning-based language model.   
     
     
         5 . The computer-implemented method of  claim 1 , wherein the target system is a machine learning based model, the computer-implemented method further comprising:
 training the machine learning based model represented by the target system using a plurality of synthetic user profiles comprising the synthetic user profile.   
     
     
         6 . The computer-implemented method of  claim 1 , wherein the target system processes information stored in user profiles, the computer-implemented method further comprising:
 performing one or more of testing or evaluation of the target system using the synthetic user profile.   
     
     
         7 . The computer-implemented method of  claim 1 , the computer-implemented method further comprising:
 generating a training dataset by repeatedly generating a new synthetic user profile by modifying the trajectory representing variation of scores of epochs while specifying a previously used set of parameters in the prompt and storing each generated user profile in the training dataset.   
     
     
         8 . The computer-implemented method of  claim 1 , wherein the trajectory representing variation of scores of epochs represents one of:
 increasing values of scores over the sequence of epochs;   decreasing values of scores over the sequence of epochs;   increasing values of scores over a first portion of the sequence of epochs and decreasing values of scores over a remaining sequence of epochs; or   decreasing values of scores over a first portion of the sequence of epochs and increasing values of scores over a remaining sequence of epochs.   
     
     
         9 . The computer-implemented method of  claim 1 , the computer-implemented method further comprising:
 specifying in the prompt, a length of time interval for each epoch of the sequence of epochs.   
     
     
         10 . The computer-implemented method of  claim 1 , the computer-implemented method further comprising:
 specifying in the prompt, a background for the hypothetical user, the background representing information about the hypothetical user based on events that occurred prior to the epochs of the sequence of epochs.   
     
     
         11 . The computer-implemented method of  claim 10 , further comprising:
 extracting the background from a set of stored backgrounds, wherein at least some of the stored backgrounds were generated using the machine learning-based language model.   
     
     
         12 . The computer-implemented method of  claim 1 , the computer-implemented method further comprising:
 specifying in the prompt:
 a first set of examples of events, wherein including an event of the first set of examples of events in an epoch causes the score of the epoch to increase, and 
 a second set of examples of events, wherein including an event of the second set of examples of events in an epoch causes the score of the epoch to decrease. 
   
     
     
         13 . A non-transitory computer readable storage medium, storing instructions that when executed by one or more computer processors cause the one or more computer processors to perform steps comprising:
 receiving a request to generate a synthetic user profile comprising a sequence of epochs, each epoch associated with a score evaluating the epoch, wherein the score is determined by executing of a set of instructions that evaluate a set of events associated with a hypothetical user that occurred during a time period, the request specifying a trajectory representing variation of scores of epochs of the sequence of epochs over time;   generating the synthetic user profile, comprising:
 generating a prompt requesting a machine learning-based language model to generate the synthetic user profile, wherein the prompt specifies a set of parameters including a representation of the trajectory specified in the request, 
 sending the prompt to the machine learning-based language model; and 
 receiving a response generated by executing the machine learning-based language model, the response comprising a sequence of synthesized epochs, wherein each synthesized epoch has synthesized events such that the scores of the sequence of synthesized epochs varies over time according to the trajectory specified in the request; and 
   providing the synthetic user profile as input to a target system.   
     
     
         14 . The non-transitory computer readable storage medium of  claim 13 , wherein the trajectory is specified as an image comprising a curve, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 specifying the image representing the trajectory as part of the prompt generated for providing to the machine learning-based language model.   
     
     
         15 . The non-transitory computer readable storage medium of  claim 13 , wherein the trajectory is specified as a sequence of tuples, each tuple comprising an x-coordinate value and a y-coordinate value, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 determining an expected score for each epoch of the sequence of epochs based on one or more of interpolation or extrapolation of values from the sequence of tuples; and   specifying the expected score for each epoch in the prompt generated for the machine learning-based language model.   
     
     
         16 . The non-transitory computer readable storage medium of  claim 13 , wherein the trajectory is specified using a natural language description of the variation of values of the scores over time, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 including the natural language description of the variation of values of the scores corresponding to the trajectory in the prompt generated for the machine learning-based language model.   
     
     
         17 . The non-transitory computer readable storage medium of  claim 13 , wherein the instructions further cause the one or more computer processors to perform steps comprising:
 generating a training dataset by repeatedly generating a new synthetic user profile by modifying the trajectory representing variation of scores of epochs while specifying a previously used set of parameters in the prompt and storing each generated user profile in the training dataset.   
     
     
         18 . A computer system comprising:
 one or more computer processors; and   a non-transitory computer readable storage medium, storing instructions that when executed by the one or more computer processors cause the one or more computer processors to perform steps comprising:
 receiving a request to generate a synthetic user profile comprising a sequence of epochs, each epoch associated with a score evaluating the epoch, wherein the score is determined by executing of a set of instructions that evaluate a set of events associated with a hypothetical user that occurred during a time period, the request specifying a trajectory representing variation of scores of epochs of the sequence of epochs over time; 
 generating the synthetic user profile, comprising:
 generating a prompt requesting a machine learning-based language model to generate the synthetic user profile, wherein the prompt specifies a set of parameters including a representation of the trajectory specified in the request, 
 sending the prompt to the machine learning-based language model; and 
 receiving a response generated by executing the machine learning-based language model, the response comprising a sequence of synthesized epochs, wherein each synthesized epoch has synthesized events such that the scores of the sequence of synthesized epochs varies over time according to the trajectory specified in the request; and 
 
 providing the synthetic user profile as input to a target system. 
   
     
     
         19 . The computer system of  claim 18 , wherein the trajectory is specified as an image comprising a curve, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 specifying the image representing the trajectory as part of the prompt generated for providing to the machine learning-based language model.   
     
     
         20 . The computer system of  claim 18 , wherein the trajectory is specified as a sequence of tuples, each tuple comprising an x-coordinate value and a y-coordinate value, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 determining an expected score for each epoch of the sequence of epochs based on one or more of interpolation or extrapolation of values from the sequence of tuples; and   specifying the expected score for each epoch in the prompt generated for the machine learning-based language model.   
     
     
         21 . The computer system of  claim 18 , wherein the trajectory is specified using a natural language description of the variation of values of the scores over time, wherein the instructions further cause the one or more computer processors to perform steps comprising:
 including the natural language description of the variation of values of the scores corresponding to the trajectory in the prompt generated for the machine learning-based language model.